How we rate sellers
Methodology version 1.1, 25 September 2026. The same rules apply to every seller. Any change gets a new version number, a changelog entry and a full re-score.
How we collect evidence
- We use only public pages: the seller's own site and independent profiles (Hellopeter, Trustpilot, Google), plus lab verification pages.
- We read pages the way a visitor does. We do not create accounts, place orders or open checkout.
- Every point we award links to the page it came from. If we cannot verify something, it scores 0 and is shown as "Not verified".
Principles
- Public, verifiable evidence only. Every point awarded must be backed by a URL (or archived copy) that any reader can check. If we cannot verify something, it scores 0 for that item and is shown as "Not verified" — not as "No".
- Same rules for everyone. No manual adjustments, no "editor's bonus".
- Snapshot-based. Each score states the date its data was collected.
- Corrections welcome. A seller can send evidence (e.g. a COA with a lab verification key). We re-check and re-score; the change is logged.
- Commercial relationships (if any) are disclosed and have no effect on scores.
Score: 100 points, six criteria
1. Lab-verified product quality — 35 pts
Why it carries the most weight: for an unregulated research compound, the main risk to the buyer is not getting what the label says (wrong compound, underdosed, impure).
| Item | Pts | How it is scored |
|---|---|---|
| 1a Published retatrutide COA from a named analytical lab | 10 | 10 = document published; 0 = none, "on request", or a claim without a document |
| 1b COA is independently verifiable | 8 | 8 = verification key printed on the COA, or a verification link to the lab's own site (we record whether we could open it; a lab's bot protection blocking us does not reduce the score); 4 = lab named and task no. shown, but the key field is empty or missing; 0 = otherwise |
| 1c COA is tied to what is sold now | 6 | 6 = batch/lot no. on COA matches product listing; 3 = batch shown but link to current stock unclear; 0 = no batch, or batch "Unknown" |
| 1d Recency | 4 | 4 = test ≤ 6 months old; 2 = 6–12 months; 0 = older / undated |
| 1e Result quality | 4 | Purity ≥ 99% and measured content within ±10% of label = 4; one of the two = 2; neither = 0. The ±10% applies in both directions: overfill counts the same as underfill, because too much compound is also a dosing risk |
| 1f Independent (buyer-submitted) test, e.g. Finnrick | 3 | 3 = passing independent test ≤ 12 months; 0 = none; −3 if a FAIL is on record |
| 1g Signs of fabricated test data | −3 | Applied when a published COA, test summary or "COA preview" shows any of: batch/lot numbers that do not exist in the seller's own lot records or on any document; results that step in a regular sequence across lots (e.g. 99.40 / 99.41 / 99.42%); a verification key the lab's site reports as invalid; content that contradicts the lab's own verified version. The finding and its evidence are shown in full |
Criterion 1 cannot go below 0.
2. Independent customer reputation — 15 pts
Only third-party platforms (Hellopeter, Trustpilot, Google Business). On-site reviews are not counted — the seller controls them.
| Item | Pts | How it is scored |
|---|---|---|
| 2a Profile exists with enough volume | 5 | ≥ 50 reviews = 5; 20–49 = 3; 5–19 = 2; < 5 = 0. Hellopeter and Trustpilot: reviews in the last 12 months. Google: total review count (Google does not show a 12-month split). If a seller has several profiles, the one with the most reviews is used |
| 2b Average rating | 6 | ≥ 4.5 = 6; 4.0–4.49 = 4; 3.5–3.99 = 2; < 3.5 = 0. Same profile and period as 2a. Only scored if 2a ≥ 2 |
| 2c Handling of complaints | 4 | Share of negative (1–2★) reviews that got a reply, on the same profile as 2a: ≥ 80% = 4; 50–79% = 2; < 50% = 0. If the profile has no negative reviews in the period, 2c scores 4 |
| Penalty | −5 | Evidence of review manipulation (e.g. hidden schema markup with fabricated reviews) |
Criterion 2 cannot go below 0.
3. Business transparency — 15 pts
| Item | Pts | How it is scored |
|---|---|---|
| 3a Legal entity name stated | 3 | |
| 3b CIPC registration number stated | 4 | 4 = number stated and found in CIPC records; 2 = stated but not verified |
| 3c Physical address | 4 | 4 = full street address, consistent across own site and third-party profiles; 2 = stated but inconsistent or suburb only; 0 = none/"on request" |
| 3d Working contact channels | 2 | 2 = two or more channels (e.g. email + phone); 1 = one |
| 3e Named responsible person | 2 |
4. Responsible marketing — 15 pts
Why it counts: retatrutide is not approved in South Africa. Sellers who make disease or weight-loss claims, or use "safe and regulated" wording, put buyers at risk and misrepresent the product's status.
Start at 15 and subtract (minimum 0):
| Finding | Deduction |
|---|---|
| Disease or weight-loss treatment claims (e.g. "fat burner", "treats diabetes") | −3 |
| Claims of regulatory approval or safety ("SAHPRA compliant", "safe and regulated") | −3 |
| Claims of testing, COAs or certifications ("every batch tested", "99%+ guaranteed", "GMP", "ISO") with no supporting document published | −2 |
| Disguised product names (e.g. "R3t@trutide", "GLP-3 (RTA)") | −2 |
| No clear research-use disclaimer | −2 |
| Using registered drug brand names (e.g. "Ozempic", "Mounjaro") for unrelated products | −2 |
5. Price transparency & value — 10 pts
| Item | Pts | How it is scored |
|---|---|---|
| 5a Prices public, with mg per unit clearly stated | 4 | 4 = all retatrutide listings; 2 = some; 0 = hidden or mg missing. The mg amount must be stated in text on the listing (title, description or options); an image alone does not count |
| 5b Price per mg vs market median (all 17 sellers, 30 mg-equivalent, cheapest in-stock listing) | 6 | ≤ 80% of median = 6; 80–100% = 4; 100–120% = 2; > 120% = 0 |
Rules for 5b:
- The price used is the one on the product page. If the same product shows a different price elsewhere on the site (cart widget, legacy page), the product page price is used and the conflict is noted.
- Only single-compound retatrutide counts, in vials or pens. Blends (e.g. with cagrilintide, tirzepatide or naltrexone) are excluded.
Price carries low weight on purpose: a cheap vial with no verified content is not good value.
6. Delivery, payment & policies — 10 pts
If policies on different pages of the same site contradict each other, the version least favourable to the buyer is scored, and the conflict is noted.
| Item | Pts | How it is scored |
|---|---|---|
| 6a Delivery terms published (courier, cost, timeframe) | 3 | 3 = all three; 1 = partial |
| 6b Cold-chain / temperature-controlled packaging stated | 2 | |
| 6c Written returns/refund policy covering wrong or damaged goods | 2 | |
| 6d Card payment through a payment gateway (buyer protection) | 3 | 3 = card payment through a gateway that is named on the site or visible in the site's public code; 1 = EFT only, or "cards accepted" with no gateway identified; 0 = crypto-only / payment details sent privately |
Tie-breaks
- Higher criterion-1 score, 2. higher criterion-2 score, 3. higher criterion-4 score. If all three are equal, the sellers share the place (shown as e.g. "11="), listed alphabetically.
Sensitivity check
After scoring, the ranking is recomputed with alternative weights (equal weights; quality 30 / reputation 20, i.e. the draft weights). If first place changes under the alternatives, the home page says the lead is not robust rather than naming a single "best" seller.
Known limitation (snapshot 2026-09-25): only two sellers have a third-party review profile with enough volume, so most sellers score 0 on criterion 2. The sensitivity check shows how much of the gap between sellers comes from reviews.
What the rating does NOT do
- It does not test products. It checks public evidence.
- It does not recommend use. Retatrutide is an investigational compound, not approved for general medical or consumer use in South Africa.
Changelog
- v1.1 (2026-09-25) — removed the deduction for human dosing instructions, schedules and injection guides on a seller's site (−6). Dosing and mixing guides are common across sellers and often generic. The change was made after the first scoring and before publication; all sellers were re-scored.
- v1.0.1 (2026-09-25) — clarifications, applied before first publication: 2c scores 4 when the profile has no negative reviews in the period; sellers still tied after all tie-breaks share the place.
- v1.0 (2026-09-25) — frozen. Changes from draft v0.1:
- Weights: criterion 1 raised from 30 to 35, criterion 2 cut from 20 to 15. Item points rescaled (1a 10, 1b 8, 1c 6; 2a 5, 2b 6, 2c 4).
- 1b: a key printed on the COA scores in full even if the lab's page is behind bot protection; an empty key field with lab and task no. scores half.
- 1c: batch "Unknown" scores 0.
- 1e: ±10% stated explicitly as applying to overfill and underfill.
- New 1g: −3 for signs of fabricated test data, with a defined list of signs.
- 2: Google counted by total reviews; 2b and 2c use the profile with the most reviews.
- 4: new −2 for testing or certification claims without a published document.
- 5a: mg must be in text, not only in an image.
- 5b: product-page price is used; blends excluded.
- 6: contradictory policies scored at the least favourable version; 6d needs an identified gateway for 3 points.
- v0.1 — internal draft, not published.
Scoring notes
These notes record how we applied the methodology to edge cases, so every seller is treated the same. They do not change the rules.
Criterion 1
- 1a counts only a document (image/PDF/page) for retatrutide from a named analytical lab that a visitor can open from the site (or its public files). "On request", "available", "tested" without a document = 0.
- 1b: key printed on COA or link to the lab's verify page = 8, even if the lab's page blocked us. Lab + task no. but empty/missing key = 4.
- 1c: batch "Unknown" = 0. If the seller sells no retatrutide now, 1c = 0.
- 1d: age measured from the newest retatrutide COA test date to 2026-09-25. ≤ 6 months = on/after 2026-03-25; 6–12 months = on/after 2025-09-25.
- 1e: use the newest retatrutide COA. Content = measured mg vs label mg, ±10% in both directions.
- 1f: Finnrick or similar buyer-submitted test for this seller.
- 1g: −3 only if one of the listed signs is documented in the report.
- Criterion 1 floor 0.
Criterion 2
- Profiles counted: Hellopeter, Trustpilot, Google Business/Maps. Use the one with the most reviews. A profile must be clearly tied to this seller (domain link, same name + matching details). Unclaimed/untied profiles = not counted.
- Trustpilot "blocked / Not verified" = treat as no profile found.
- 2c: no negative (1–2★) reviews in the period = 4 (nothing left unanswered). Note any clustering of reviews in flags.
- Penalty −5 for documented manipulation, e.g. hidden schema.org reviews with templated text, or review markup on products that cannot have been bought. Record it even if the criterion is already 0 (floor 0).
Criterion 3
- 3a/3b: an entity name or CIPC number that appears anywhere on the site, including hidden structured data (JSON-LD), counts as "stated". Note "hidden markup only" where that applies. CIPC not checked against CIPC records = 2 (stated but not verified). Two different entity names on one site still count for 3a; note the conflict.
- 3c: full street address = 4 only if consistent; suburb/city only = 2; "on request" / "South Africa" / region only = 0.
- 3d channels: email, phone/WhatsApp (same number counts once), contact form. Social media accounts and hidden (not visible) emails do not count.
- 3e: a named person responsible for the business (founder/owner/director).
Criterion 4 (start 15, floor 0)
- Weight-loss/disease claims (−3): buyer-directed benefit wording such as "weight loss", "fat burning", "weight management", "treats diabetes", "Fat-Loss" category, weight-loss testimonials shown by the seller.
- Approval/safety (−3): "SAHPRA compliant", "safe", "regulated", "approved".
- Unsubstantiated testing/certification (−2): "every batch tested", "third-party tested", "99%+ guaranteed", "GMP", "ISO", "public COA on every batch" while no retatrutide COA document is published. If a real retatrutide COA is published, no deduction for general testing claims.
- Disguised names (−2): the listing does not name retatrutide plainly ("Reta 30", "GLP-3 (RTA)", "R E T A", "RETA 30mg" with retatrutide only in small print still counts as disguised if the product name itself hides it).
- No clear disclaimer (−2): no research-use-only statement on the retatrutide page or site-wide footer/terms.
- Drug brand names (−2): products named or tagged with Ozempic, Mounjaro, Wegovy, Zepbound, "iZempic", "Ozempic Alternative" etc. Mentions in an educational comparison table do not count.
Criterion 5
- 5a: mg in text on every pure retatrutide listing = 4; some = 2; none/hidden = 0.
- 5b: among in-stock pure retatrutide listings (vials or pens; no blends; backorder/sold out excluded), take the one with the lowest price per mg, using the product-page price, and compare it with the median of all rated sellers. If there is no such listing, or the mg amount cannot be read from the product page, 5b scores 0.
Criterion 6
- Contradictory policies: score the version least favourable to the buyer.
- 6a: courier named + cost + timeframe = 3; any one or two = 1; none = 0.
- 6b: temperature-controlled / cold-chain / ice packs stated = 2.
- 6c: written policy that covers wrong or damaged goods = 2 (a 48-hour reporting window still counts). No policy, or policy excludes it = 0.
- 6d: gateway named on site or visible in public code/API = 3; EFT only, or "cards accepted" with no gateway identified = 1; crypto-only / private = 0. EFT + crypto with no gateway = 1.